Add Unitxt Multimodality Support (#2364)

* Add Unitxt Multimodality Support
Signed-off-by: elronbandel <elronbandel@gmail.com>

* Update
Signed-off-by: elronbandel <elronbandel@gmail.com>

* Fix formatting
Signed-off-by: elronbandel <elronbandel@gmail.com>

---------
Signed-off-by: elronbandel <elronbandel@gmail.com>

.gitignore

@@ -8,6 +8,7 @@ build
 dist
 *.egg-info
 venv
+.venv/
 .vscode/
 temp
 __pycache__

lm_eval/tasks/unitxt/README.md

 # Unitxt
 
+Unitxt is a library for customizable textual data preparation and evaluation tailored to generative language models. Unitxt natively integrates with common libraries like HuggingFace and LM-eval-harness and deconstructs processing flows into modular components, enabling easy customization and sharing between practitioners. These components encompass model-specific formats, task prompts, and many other comprehensive dataset processing definitions. These components are centralized in the Unitxt-Catalog, thus fostering collaboration and exploration in modern textual data workflows.
+
+The full Unitxt catalog can be viewed in an [online explorer](https://unitxt.readthedocs.io/en/latest/docs/demo.html).
+
+Read more about Unitxt at [www.unitxt.ai](https://www.unitxt.ai/).
+
 ### Paper
 
 Title: `Unitxt: Flexible, Shareable and Reusable Data Preparation and Evaluation for Generative AI`
-Abstract: `https://arxiv.org/abs/2401.14019`
-
-Unitxt is a library for customizable textual data preparation and evaluation tailored to generative language models. Unitxt natively integrates with common libraries like HuggingFace and LM-eval-harness and deconstructs processing flows into modular components, enabling easy customization and sharing between practitioners. These components encompass model-specific formats, task prompts, and many other comprehensive dataset processing definitions. These components are centralized in the Unitxt-Catalog, thus fostering collaboration and exploration in modern textual data workflows.
+Abstract: [link](https://arxiv.org/abs/2401.14019)
 
-The full Unitxt catalog can be viewed in an online explorer. `https://unitxt.readthedocs.io/en/latest/docs/demo.html`
 
-Homepage: https://unitxt.readthedocs.io/en/latest/index.html
 
 ### Citation
 
@@ -36,42 +38,4 @@ The full list of Unitxt tasks currently supported can be seen under `tasks/unitx
 
 ### Adding tasks
 
-You can add additional tasks from the Unitxt catalog by generating new LM-Eval yaml files for these datasets.
-
-The Unitxt task yaml files are generated via the `generate_yamls.py` script in the `tasks/unitxt` directory.
-
-To add a yaml file for an existing dataset Unitxt which is not yet in LM-Eval:
-1. Add the card name to the `unitxt_datasets`  file in the `tasks/unitxt` directory.  
-2. The generate_yaml.py contains the default Unitxt [template](https://unitxt.readthedocs.io/en/latest/docs/adding_template.html) used for each kind of NLP task in the `default_template_per_task` dictionary.  If the dataset is of a Unitxt task type, previously not used in LM-Eval, you will need to add a default template for it in the dictionary.  
-
-```
-default_template_per_task = {
-     "tasks.classification.multi_label" : "templates.classification.multi_label.title" ,
-     "tasks.classification.multi_class" : "templates.classification.multi_class.title" ,
-     "tasks.summarization.abstractive" :  "templates.summarization.abstractive.full",
-     "tasks.regression.two_texts" : "templates.regression.two_texts.simple",
-     "tasks.qa.with_context.extractive" : "templates.qa.with_context.simple",
-     "tasks.grammatical_error_correction" : "templates.grammatical_error_correction.simple",
-     "tasks.span_labeling.extraction" : "templates.span_labeling.extraction.title"
-}
-```
-3. Run `python generate_yaml.py` (this will generate all the datasets listed in the `unitxt_datasets`)
-
-If you want to add a new dataset to the Unitxt catalog, see the Unitxt documentation:
-
-https://unitxt.readthedocs.io/en/latest/docs/adding_dataset.html
-
-
-
-### Checklist
-
-For adding novel benchmarks/datasets to the library:
-* [ ] Is the task an existing benchmark in the literature?
-  * [ ] Have you referenced the original paper that introduced the task?
-  * [ ] If yes, does the original paper provide a reference implementation? If so, have you checked against the reference implementation and documented how to run such a test?
-
-
-If other tasks on this dataset are already supported:
-* [ ] Is the "Main" variant of this task clearly denoted?
-* [ ] Have you provided a short sentence in a README on what each new variant adds / evaluates?
-* [ ] Have you noted which, if any, published evaluation setups are matched by this variant?
+See the [adding tasks guide](https://www.unitxt.ai/en/latest/docs/lm_eval.html#).

lm_eval/tasks/unitxt/doc_vqa.yaml

+task: doc_vqa
+include: unitxt_multimodal
+recipe: card=cards.doc_vqa.en,template=templates.qa.with_context.title

lm_eval/tasks/unitxt/task.py

@@ -4,9 +4,12 @@ In the dynamic landscape of generative NLP, traditional text processing pipeline
 Addressing this need, we present Unitxt, an innovative library for customizable textual data preparation and evaluation tailored to generative language models. Unitxt natively integrates with common libraries like HuggingFace and LM-eval-harness and deconstructs processing flows into modular components, enabling easy customization and sharing between practitioners. These components encompass model-specific formats, task prompts, and many other comprehensive dataset processing definitions. The Unitxt-Catalog centralizes these components, fostering collaboration and exploration in modern textual data workflows. Beyond being a tool, Unitxt is a community-driven platform, empowering users to build, share, and advance their pipelines collaboratively.
 """
 
+import importlib.util
+import re
 from functools import partial
-from typing import Optional
+from typing import Any, Dict, Optional
 
+import datasets
 import evaluate
 
 from lm_eval.api.instance import Instance
@@ -25,6 +28,10 @@ _CITATION = """
 """
 
 
+def is_unitxt_installed() -> bool:
+    return importlib.util.find_spec("unitxt") is not None
+
+
 def score(items, metric):
     predictions, references = zip(*items)
     evaluator = evaluate.load("unitxt/metric")
@@ -41,17 +48,30 @@ class Unitxt(ConfigurableTask):
         self,
         config: Optional[dict] = None,
     ) -> None:
+        if config is None:
+            config = {}
         assert "recipe" in config, "Unitxt task must have a 'recipe' string."
         super().__init__(
             config={
                 "metadata": {"version": self.VERSION},
-                "dataset_kwargs": {"trust_remote_code": True},
                 "dataset_name": config["recipe"],
-                "dataset_path": "unitxt/data",
             }
         )
+        self.image_decoder = datasets.Image()
         self.metrics = self.dataset["test"][0]["metrics"]
 
+    def download(self, dataset_kwargs: Optional[Dict[str, Any]] = None) -> None:
+        if is_unitxt_installed():
+            from unitxt import load_dataset
+
+            self.dataset = load_dataset(self.DATASET_NAME)
+        else:
+            self.dataset = datasets.load_dataset(
+                name=self.DATASET_NAME,
+                path="unitxt/data",
+                trust_remote_code=True,
+            )
+
     def has_training_docs(self):
         return "train" in self.dataset
 
@@ -79,6 +99,9 @@ class Unitxt(ConfigurableTask):
     def doc_to_target(self, doc):
         doc["target"]
 
+    def get_arguments(self, doc, ctx):
+        return (ctx, {"until": ["\n"]})
+
     def construct_requests(self, doc, ctx, **kwargs):
         """Uses RequestFactory to construct Requests and returns an iterable of
         Requests which will be sent to the LM.
@@ -90,12 +113,11 @@ class Unitxt(ConfigurableTask):
             language description, as well as the few shot examples, and the question
             part of the document for `doc`.
         """
-
         return [
             Instance(
                 request_type="generate_until",
                 doc=doc,
-                arguments=(ctx, {"until": ["\n"]}),
+                arguments=self.get_arguments(doc, ctx),
                 idx=0,
                 **kwargs,
             )
@@ -140,3 +162,34 @@ class Unitxt(ConfigurableTask):
             whether a higher value of the submetric is better
         """
         return {metric.replace("metrics.", ""): True for metric in self.metrics}
+
+
+images_regex = r'<img\s+src=["\'](.*?)["\']\s*/?>'
+image_source_regex = r'<img\s+src=["\'](.*?)["\']'
+
+
+def extract_images(text, instance):
+    image_sources = re.findall(image_source_regex, text)
+    images = []
+    for image_source in image_sources:
+        current = instance
+        for key in image_source.split("/"):
+            if key.isdigit():
+                key = int(key)
+            current = current[key]
+        images.append(current)
+    return images
+
+
+class UnitxtMultiModal(Unitxt):
+    MULTIMODAL = True
+
+    def doc_to_text(self, doc):
+        return re.sub(images_regex, "<image>", doc["source"])
+
+    def doc_to_image(self, doc):
+        images = extract_images(doc["source"], doc)
+        return [self.image_decoder.decode_example(image) for image in images]
+
+    def get_arguments(self, doc, ctx):
+        return (ctx, {"until": ["\n"]}, {"visual": self.doc_to_image(doc)})

lm_eval/tasks/unitxt/unitxt_multimodal

+class: !function task.UnitxtMultiModal